What Software Should I Learn? Pt. I

Linux and Shell Scripting

If you are preparing to enter graduate school in the sciences, or have recently started a graduate program, you have probably become keenly aware that working with modern datasets requires the use of tools that are more powerful and more esoteric than Excel. One way or another, you will have to learn how to code, and it probably behooves you to learn this skill sooner rather than later. You might start off graduate school like me, having taken one intro to computer science class as an undergraduate, where I learned to write rudimentary programs in C. On the other end of the spectrum, you may come to graduate school with a fair amount of coding experience under your belt, looking to refine your skills and learn how to apply them to your chosen research subject.

Whatever the case, I think it is helpful to review the various applications for coding that you’re likely to run into in graduate school. My advice here is geared towards the life sciences, as my field of study falls within this domain. I won’t offer any specific advice on how exactly to learn a new language, because there isn’t any magical, easy way to go about it. You should learn how to code the same way you learn everything in school – by taking a class, either in-person or online. More importantly, during or after that class you should then learn the new coding language for real by working on a project that requires you to use it. At first you will stumble about, writing terrible code that gives you grotesque, misshapen output. Eventually you’ll progress to writing terrible code that gives you the correct output. Someday down the road you may write very good code that other people use in their research. Otherwise you’ll be like me, writing mediocre code that gets the job done. Either way, you will have learned a valuable skill that will make your future academic life easier.

What is Linux?

If your research project(s) will involve any degree of bioinformatic work, in particular the processing of sequencing data, learning how to use Linux will be absolutely essential. It’s very unlikely that you haven’t heard of Linux, but if that is the case, Linux is one of several “Unix-like” operating systems. Also included among the ranks of Unix-like operating systems is one you may not have heard of, BSD, and one you have heard of – MacOS. The Unix-like OSes are, of course, like Unix, which is an operating system that was first developed at Bell Labs starting in the 1970s. Linux itself is named after its developer, Linus Torvalds, and was first released in 1991. Since then, many thousands of different versions, or distributions of Linux have been developed. Whatever distribution you choose (more on that below), you can open a terminal (actually technically a terminal emulator) and begin typing in commands by hand. The interface between the user and the operating system is called a shell. There are several different shells out there – sh, zsh, fish, and the most popular, bash. The magic of Linux and other Unix-like operating systems is that the commands that you would normally type out by hand can be combined into scripts – shell scripts – to automate the process.

What should I use the Linux shell for?

Why bother with learning how to use Linux, when Windows and MacOS dominate the personal computer market? In your mind, Linux and shell scripting should be associated with one type of data – large text files. This speaks both to the strengths and weaknesses of the shell. You see, the shell’s greatest weakness is that it treats everything it sees as strings (i.e. text). That means that if you type in something that you think of as a number, like “12,” the shell will treat it as text, unless you specifically tell it to interpret it as a number using some syntax that is frankly rather clunky. On the other hand, the shell, along with all its associated utilities, does its job of text processing very, very well.

It just so happens that many of the files containing biological data that you encounter will be large text files. This isn’t by coincidence. These include:

• Files of unaligned sequences, either with associated quality information (FASTQ) or without (FASTA). These are typically referred to as “formats” or “specifications,” but that is technically incorrect. There is no formal specification for how these files should be formatted. Rather, they represent sets of informal conventions that have been adopted by the community over time.
• Sequence alignment files (SAM). Typically you would generate a SAM file by aligning the reads in a FASTQ or FASTA file against a reference sequence.
• Genetic variant data stored in variant call format (VCF) files. These are typically generated by comparing multiple sequence alignment files and seeing where they differ.

You may have already encountered these types of files. They usually are compressed (often with the extension “.gz”), and if you try to open one with a text editor on your typical laptop it may promptly crash. Why does this happen? Because the file when uncompressed could easily be larger than the amount of memory on your computer. For instance, you may encounter a 4GB file with the extension “.fastq.gz.” However, when uncompressed this file could actually be 15GB. If you try to open it on a computer with 8GB of RAM, it will completely fill all available memory, bringing everything to a standstill. In Linux, many tools avoid this problem by streaming – reading an input file one line at a time, doing some operation on each line, and then writing it to an output file. Since these utilities only hold one line in memory at a time, they can process files of arbitrary size. It’s important to note that they are also quite fast – likely far faster than anything you could write yourself, since they have been developed over the course of several decades. The primary limiting factor for file operations becomes the read/write speed of the storage medium.

The ability of most Linux command-line utilities to handle very large files is one reason that the large computers you will encounter in your research – servers and high performance computing clusters (i.e. supercomputers) – will almost invariably run some kind of Linux distribution. They will also almost invariably have no graphical user interface. So when I talk about “learning how to use Linux,” I specifically mean learning how to work with the command line. If your experience is like that of many thousands of graduate students around the world, you may get some general guidance from your university’s IT department on how to log in to some particular server using a secure shell (ssh) interface. Upon logging in, you will be confronted by an empty black screen with a blinking cursor. What now?

Some Simple Examples

Linux, like Unix, uses the input redirect (<), pipe (|), and output redirect (>) operators. Most command line utilities will by default read from standard input (the keyboard) and write to standard output (the screen), but these operators allow you to read input from a file, use the output of one command as the input of another, and send output to a file instead of the screen. The input redirect operator is used less frequently than the other two, because many command line utilities can be pointed directly to an input file to read from. For instance, say we had a gzip-compressed fasta file in the typical format of sequence IDs preceded by “>”, sequence descriptions appearing after the first space in ID lines, and sequences as everything not proceeded by “>”. For instance:

>sequence1_id sequence1_description
AGTCCGCATAG
>sequence2_id sequence2_description
GGTCACTTTGA
>sequence3_id sequence3_description
TTACTGGCAAA
...

If we wanted to, say, isolate the first 10,000 sequence IDs (without the “>” preceding them) and place them into a text file, we could write:

gunzip -c my_fasta_file.fa.gz |
    grep ">" -m 10000 |
    sed 's/>//' |
    sed 's/ .*$//' > my_fasta_ids.txt

Here we perform the following steps, using pipes to supply the output of one step as the input for the next:

1. decompress the gzipped fasta file using gunzip -c. The “-c” option stands for “console,” and would normally write the output to the screen, except here we redirect using the pipe operator
2. Use grep, which pulls out any line containing “>”. Using the -m (for max) option limits the output to the first 10,000 matches.
3. Use the Unix stream editor program (sed) to reformat our sequence ID lines. Here we are using find/replace commands, written in the strange-looking nomenclature ‘s/find/replace/’. Here “s” stands for “string,” so in the first line we’re replacing the string “>” with nothing (that is, just removing it).
4. Perform a second sed replacement using a regular expression to replace everything after the first space (including the space) with nothing.
5. Redirect our final output to my_fasta_ids.txt.

Notably, you can use this short command to isolate the first 10,000 sequence IDs from a fasta file containing 100,000 sequences, or 100,000,000 (as long as it can fit on your hard drive). Some, but not all Linux systems will have the program zgrep installed, which is a version of grep that can directly read gzip-compressed files. In addition, we can supply sed with multiple find/replace expressions by using the -e option multiple times, so we could condense the above command to:

zgrep ">" -m 10000 my_fasta.fa.gz |
    sed -e 's/>//' -e 's/ .*$//' > my_fasta_ids.txt

Many bioinformatics softwares are designed to process specific types of the files listed above, using syntax that is more or identical to general shell syntax. So we have the program samtools for working with SAM files, and the program bcftools (or the older vcftools) for working with VCF files. We can do various piping/redirection operations using these, so for instance to output the chromosome, position, reference allele and alternate allele for all SNPs on chromosome 2 in a given VCF file, we could write:

bcftools view my_input.vcf.gz --regions chr2 |
    bcftools query - '%CHROM\t%POS\t%REF\t%ALT\n' > chr2_snp_info.txt

Here we are using bcftools view to isolate just chromosome 2, and then passing that off to bcftools query (the lonesome “-” here means “read from standard input,” i.e. the pipe’s output). Then we have a very literal statement showing bcftools query how to write the output, separating our desired output fields by tabs (“\t“) and signifying that we want each SNP to be listed on a separate line (“\n“).

Hardware for Learning Linux

You always have the option of learning Linux without any graphical interface, as you would on a university server or cluster, or else through a virtual machine. However, it can be a hassle to have to deal with these systems before you have the basics down. If you want to go a more local route, you have a few options:

1. A PC running a Linux distribution. This is by far the best option. There are a huge amount of distributions to choose from – some popular ones are Ubuntu (though it faces some criticism due to corporate practices of its parent company Canonical), Linux Mint (a very popular community-driven fork of Ubuntu), Debian, Zorin OS and Elementary OS (both good for people used to Windows or Mac), Manjaro, Gentoo, and the list goes on. I personally prefer Linux Mint. Working on a Linux computer will allow you to try out various commands and scripts on small test datasets before running larger jobs on servers or clusters. It’s also much easier to learn how to use the command line locally, where you don’t have to worry about constantly logging in to a remote computer, making other users mad by hogging resources, or else scheduling jobs like you would on a cluster. It’s also probably easier than you think to obtain a computer running Linux. If you have an old laptop or desktop lying around that struggles to run Windows or MacOS, you can often install a lightweight Linux distribution and magically resurrect it. Some good distributions to consider in this case would be those using the Xfce desktop environment, such as Linux Mint Xfce edition, or those using the LXDE desktop environment, such as Lubuntu.
2. A dual-boot computer. This is when you have two operating systems installed side-by-side on the same machine. This way you can have both Windows or MacOS installed alongside a Linux OS. This might seem like the best of both worlds, but there are some caveats. Switching between the two OSes requires restarting the machine, which can become tedious. Also, setting up dual boot will immediately void your warranty if it is still active.
3. Use Windows 10 or MacOS. Traditionally, Macs were considered superior to Windows PCs if you wanted a popular, commercial operating system that somewhat resembled Linux. Recall that both Linux and MacOS are Unix-like. However, MacOS more closely resembles BSD than Linux, and it requires some tweaking to truly get it working more like Linux. Even then, it’s never quite the same. By comparison, Windows was the odd one out, excepting a few solutions like Cygwin or virtual machines (VMs). However, that changed somewhat with the release of Windows 10 and the Windows Subsystem for Linux (WSL). This is a VM that allows you to install one of several Linux distributions inside of Windows. The first version of WSL was quite a bit faster than a traditional VM, though still somewhat clunky. The new WSL 2 has vastly improved performance over the original WSL. However, there is a catch – WSL runs with its own filesystem, and it can be quite slow when working with anything in the Windows filesystem on the computer. With the release of WSL 2, Windows has become more competitive with Mac when it comes to Linux functionality. However, if you’re looking for a seamless experience for learning Linux, Both Windows and Mac will still give you some headaches from time to time.

Whatever route you choose, the first bit of learning might be a little bit tough and frustrating. However, now that you can look up or ask any question on the internet, the task of learning Linux has become a lot easier. Soon enough you’ll be breezing through the command line and processing enormous files, surprising your advisor by getting through huge computational projects with relative ease.

Misconceptions About Genetics and Evolution in Pop Culture, Part…

Mutation

For part II of this series, we’ll tackle the big one – perhaps one of the most misunderstood natural phenomena in all of fiction. In writing, movies, television shows, video games and countless other media, genetic mutation is often viewed as something frightening or uncontrollable, frequently brought about through the abuse or misuse of science, or else through unfortunate accidents. For instance, genetic mutation turns the physicist Bruce Banner into the Hulk – a violent, simple-minded brute whose favorite hobby is smashing things.

In reality, mutation is neither a good thing nor a bad thing. It is simply a natural process, like the weather or plate tectonics. Mutation is the raw fuel for natural selection and evolution. It is the spontaneous change in DNA sequence over generations due to either environmental factors (e.g. exposure to ionizing radiation or mutagenic chemicals) or DNA replication errors. This last part is important – mutations would occur in a population over time even if it were magically shielded from any environmental factors that cause mutation. Most small-scale mutations are single nucleotide polymorphisms (SNPs), where a single nucleotide is changed, for instance changing the sequence ATTGCTCCA to ATTGATCCA. This can also include deletions: ATTGCTCCA to ATTGTCCA, and insertions: ATTGCTCCA to ATTGCATCCA. However, larger-scale mutations such as the deletion of entire genes are also possible.

How Does Fiction Get it Wrong?

In general, fictional accounts of genetic mutation fail to grasp one major concept: the difference between somatic mutations and germline mutations. Germline cells are the small number of cells in your body that are involved in passing your genes on to your progeny. They include eggs, sperm, the cells of the ovaries involved in egg formation (oogenesis), and the cells of the testicles involved in sperm formation (spermatogenesis). Because these cells pass genetic information on to offspring, mutations arising in them will potentially be passed on to every cell of an offspring’s body as it forms through a complex series of cell divisions following the fusion of an egg and sperm. In general, female mammals are born with all of their eggs already made (there is some debate on this though). Therefore mutations to the cells involved in oogenesis after birth will not be passed on to a woman’s offspring. This is not the case in males, where the cells involved in spermatogenesis are busy creating sperm throughout most of their life.

In contrast to germline cells, somatic cells make up nearly the entire mass of your body. While some are involved in the act of reproduction (e.g. the uterus in women, or the urethra in males), the defining characteristic of somatic cells is that they do not pass any genetic information on to an organism’s progeny. This brings us to the primary error regarding mutation in fiction – mutations to somatic cells cannot effect an organism’s body as a whole. For instance, a mutation occurring in a skin cell on your forearm due to ultraviolet light exposure is unlikely to give you any superpowers. In contrast, an organism can acquire a mutation in a germline cell, which is then passed on to all cells in their offspring. These two distinct outcomes of mutation continue to get confused in science fiction. Below we’ll shed a little more light on the difference between the two.

The Effects of Somatic Mutation

Usually, mutations in somatic cells have no effect. Your genome contains a large amount of genetic material that doesn’t directly affect your phenotype. This includes repetitive elements – somewhat autonomous DNA sequences that either “copy and paste” or “cut and paste” themselves across the genome. It also includes genes that have been epigenetically silenced, and pseudogenes, which are genes that have lost their function (due to prior mutations). Mutations within any of these regions, and many others, will typically not have any measurable effect on a cell’s phenotype.

But what if a mutation occurs inside a gene?

Then things can get a little more interesting. There’s still a very good chance that a mutation inside of a gene will have little or no effect. Genes are first transcribed, when an RNA copy of the DNA sequence is created. Then the RNA transcript is translated into a protein (which is a chain of amino acids linked together). However, after transcription and before translation, portions of the RNA are removed. Since these regions, called introns, don’t make it to the final protein product of the gene, mutations within them are often “silent.”

But what if a mutation occurs in a coding region in a gene?

The regions of genes that are transcribed in RNA and then translated into proteins are called exons. Even if a mutation occurs in one of these regions, it may not have any effect. For one thing, the genetic code is degenerate. This doesn’t mean that that it’s out making trouble (usually). Rather, the genetic code consists of three-letter “words,” called codons. Each codon is translated into an amino acid – the building blocks of proteins mentioned above. So, for instance, the three codons CTTGTTGGC will be translated into the amino acid chain Leucine-Valine-Glycine. However, nearly all the amino acids are encoded by multiple codons. For instance, glycine is encoded by the codons GGC, GGT, GGA, and GGG, so any substitution of the last base of a codon starting with “GG” will not have any effect on the protein that is translated from the gene. This is what is referred to as a “synonymous” mutation.

But what if a coding region mutation isn’t synonymous?

Now things start to get more interesting. A mutation that results in a change to an amino acid is referred to as a “missense” mutation. These mutations can have little or no effect, or major effects depending on how the amino acid swap changes the overall shape or function of the protein. In addition, there are two types of mutations that can have more dramatic effects. One is a premature stop formation. There are three codons that signal for transciption to stop – TAG, TAA, and TGA. If a mutation converts, for instance, a cysteine codon (TGC) into the stop codon TGA, then it will result in a shortened protein, which can have highly significant effects. This is referred to as a “nonsense” mutation. Finally, the other more serious type of mutation is a frame shift. This occurs when an insertion or deletion of one or more bases completely changes the “words” (codons) of the genetic code in a coding sequence. To illustrate, here’s an English sentence that happens to consist of three-letter words, like the genetic code: “The fat cat and big dog ate.” If we insert an adenine (A) at the beginning of the sentence, but keep the word separations at the same place, we end up with: “Ath efa tca tan dbi gdo gat e.” Basically, a frameshift mutation scrambles codons, resulting in the production of a completely different protein.

Mutations with Major Effects

As we have seen, mutations often have no beneficial or detrimental effect. In fact, the neutral theory of molecular evolution, proposed by Motoo Kimura in 1968, posits that the majority of evolution on the molecular scale is due to the genetic drift (the random change in allele frequencies over time) of mutations with neutral effects. However, there are cases where mutations can have major phenotypic effects. Germline mutations with major effects may be enriched in or purged from a population through natural selection. Highly deleterious mutations may be lethal to organisms, while highly beneficial mutations may grant an organism greater fitness, increasing the likelihood that the mutation will be passed on to future generations.

In somatic cells, we only really see the outcome of deleterious mutations. The cells making up your body must all “play by the rules.” Specifically, they have to behave as part of a multicellular organism. Mutations that allow a cell to function beyond its very narrowly-regulated niche within the body can have very dangerous consequences. Cells have many contingencies to deal with such situations. Apoptosis is programmed cell death, whereby a cell is triggered, either internally or externally, to self-destruct. Many cells within your body undergo this process every day. There are many ways in which apoptosis is initiated; one way is in response to mutations that alter how a cell grows and multiplies. In this way, cells that pick up mutations that could allow them to stop playing nice with the other cells around them often automatically self-destruct. Unfortunately these mechanisms don’t always work. Sometimes the very pathways meant to induce apoptosis become corrupted. Unfortunately these changes don’t give us new superpowers – instead they may lead to the formation of cancerous cells.

Who Gets it Right

Given all the intricacies of how mutation works in reality, different works of fiction fall close to or wide of the mark. In the Marvel Universe, mutants are those born with mutations (presumably acquired as germline mutations by their parents or other ancestors), whereas mutates are characters who acquire their powers from some external force, energy, catastrophe, etc. So, comics that focus mostly on mutants, such as the X-Men, largely get the concept of genetic mutation correct. It’s still not clear how mutations seem to give everyone big muscles or big breasts. Nor is it clear how they occasionally give people god-like powers. In contrast, comics about mutates, such as The Hulk, Spiderman, or the Fantastic Four, have a less realistic depiction of the effects of mutations (relatively speaking), as we are supposed to believe that the exposure of a fully-developed human to some mutagen could alter their entire body. As we have seen, the most dramatic effect on somatic cells due to exposure to mutagens is cancer. This isn’t quite as exciting a subject as heroes fighting monsters, though there is a genre of comics and graphic novels dealing with illness and medicine. Interestingly, there are some cases in the Marvel Universe where mutates later give birth to mutants. One example are the mutates Sue Storm and Reed Richards of the Fantastic Four, who give birth to the mutant Franklin Richards. It turns out Franklin is essentially a god – he can alter reality itself and create universes. It’s not clear how the Marvel Multiverse still exists at all when children wield this much power.

Who Gets it Wrong

There are too many to list here – take your pick really. We already mentioned the mutates in the Marvel Universe. We could add to that an almost unlimited number of fictional characters, including many of the characters and monsters from the Resident Evil video game series, the supermutants from the Fallout video games, Joker from the Batman comic books, the Teenage Mutant Ninja Turtles – the list goes on and on.

Room for Artistic License

There is one potential mutagen that could more feasibly create mutations in a wide range of somatic cells – viruses, specifically retroviruses. These are viruses with RNA genomes which are reverse-transcribed into DNA copies, which are then inserted into the host cell genome. In this way, retroviruses naturally genetically modify their host cells. For this reason, retroviruses are the vectors of choice for gene therapy. Therefore it’s not too hard to imagine a scenario in which gene therapy is used not just to cure disease, but to grant some enhancement or even powers to humans. This ignores the many hurdles that have made successful gene therapy difficult to achieve, but it is at least within the realm of possibility. Interestingly, the human genome contains many inactive endogenous retroviruses. These were retroviruses which at one time infected germline cells in our ancestors, and have since been passed down through the generations. While retroviruses are a (somewhat) plausible method of creating body-wide mutations in an adult, we first need much more research into using them to cure disease, before we start using them to give people rockin’ bods and superpowers.

New Wheat GWAS Publication

As usual, I’ve been remiss in updating this site. However, this time I have a good reason for it. I am the first author on two journal articles that were just published within a day of each other at the end of last week. These articles both used the same data collected from the field, consisting of around 320 wheat lines grown in two locations across two years, but they encompass quite different analyses. A common theme that I’ve picked up when talking to friends and family is that nobody knows what I do for work, nor why I do it. In that spirit, I wanted to give some less-technical background on these papers here – if you want the nitty-gritty, both articles are available open-access. I’ll just go over one of them for now, and will save the other one for a future post.

Definitions

Before starting, we’ll have to define a few things, so that everything that follows will make sense. An individual’s phenotype is what you can observe about them. We typically divide phenotype into specific traits. In the case of plants these might be height, grain weight, branching, flowering time, etc. An individual’s genotype is their genetic makeup. Genotype contributes to phenotype, in conjunction with environment (an individual’s surroundings). Finally, an organism’s genome is composed of a set of very long molecules called chromosomes, each one of these being the familiar double helix consisting of a chain of bases or nucleotides, which form an organism’s “genetic code”. A marker is some specific genomic feature that we can use to differentiate between different lines. In this case, the markers we’re referring to are single nucleotide polymorphisms (SNPs), which is when a single nucleotide is substituted for another one.

Here we will be talking about bread wheat, which is the product of ancient hybridizations between multiple grass species in the Middle East. This means that it confusingly has three genomes inherited from three separate progenitor species, labeled A, B, and D. The A genome was inherited from red wild einkorn wheat (Triticum urartu). The B genome was likely inherited from a close relative of the grass species Aegilops speltoides, while the D genome was inherited from Tausch’s goat grass (Aegilops tauschii). Each genome has seven chromosomes, labeled 1A through 7A, 1B through 7B, and 1D through 7D, for a total of 21 chromosomes. The separate genomes are sometimes referred to as “subgenomes,” but here we will just refer to them collectively as “the genome.” A very common question is “where is the C genome?” The C genome is present in another Middle Eastern grass species, Aegilops caudata, which is not a progenitor species of modern wheat. I suspect that this is a historical quirk due to taxonomic reclassifications, though it’s not something I’ve looked into in great detail.

Finding marker-trait associations

In this first paper, published in PLOS One, my coauthors and I performed a genome-wide association study of various yield-related traits. What does that mean? A genome-wide association study, or GWAS, is a type of study in which we genotype all the included lines with a set of markers spread throughout the genome. Then we determine which markers (if any) explain a significant portion of the phenotypic variation that we observe in the field, that is, which markers are involved in significant marker-trait associations. In our case, we used tens of thousands of markers, though in many other species with more advanced genomic data available (especially in humans), it’s customary to use millions or tens of millions of markers.

When I say “yield-related traits,” I’m referring both to grain yield, the most important trait in cereals, as well as traits like grain production per unit area, grains per head, heads per unit area, etc. We also included several phenological traits (i.e. traits relating to the timing of plant development), such as heading date, plant maturity date, and the period of time between the two.

GWAS studies are a dime a dozen these days. What I think makes this one interesting is that it is one of the first wheat GWAS to use a reference genome. When I started graduate school, there was no reference genome, and nobody thought that there would be one anytime soon. Soon afterwards, the International Wheat Genome Sequencing Consortium sequenced the largest wheat chromosome, 3B. Weighing in at about 1 billion bases, chromosome 3B is more than twice as large as the entire rice genome. The full wheat genome is enormous, consisting of somewhere around 15 billion bases – over five times as large as the human genome. To make matters worse, much of this DNA (about 85%) consists of repetetive elements – quasi-autonomous sequences which “copy and paste” themselves throughout the genome. I personally find these sequences fascinating, but they do greatly complicate the task of assembling sequencer reads together. The monumental task of sequencing chromosome 3B convinced many researchers that finishing the entire genome could take a very long time. However, the rapid advancement of assembly technologies brought us a surprise – a fully assembled reference genome published last year.

Candidate genes

Having a reference genome available is a real game-changer. Previously, we had to rely on genetic positions for markers. These have the unfortunate property of being variable depending on the population that was used to calculate them. With a reference genome, we can locate markers both by their genetic position, and by their physical position on the chromosome. This makes it much easier to compare significant results between studies. In addition, once we have found a significant marker, we can now more easily look for candidate genes (i.e. genes that we think are causing the significant marker-trait association) surrounding it.

Sometimes we get lucky, and a particular marker is itself causing variation within a gene, which is in turn causing phenotypic variation that we can observe. However, more often that not, a marker is merely floating out between genes somewhere, presumably close to one that we are actually interested in. In this case, the alleles of the marker and different alleles of the gene will be held together in linkage disequilibrium (LD) or gametic phase disequilibrium. These are two complicated terms for the same, somewhat simple concept. If you learned about Gregor Mendel’s peas and Punnett squares in high school biology, you likely learned about Mendel’s Law of Independent Assortment, which states that alleles (i.e. different “versions”) of separate genes are inherited independently. This is often true, but not always. Specifically, if two genes, or a gene and a marker, are located close to each other on a chromosome, their alleles will be in LD, meaning they will be inherited together in a non-random fashion. There are other factors that can cause genes to be in LD with each other, but physical proximity is the most prevalent. Using the reference genome, we could finally look at patterns of LD in terms of physical distances. In the figure below, we have LD plotted against physical distance, both in the A, B, and D genomes as a whole (top panel), and in each of the seven chromosomes within each genome (bottom panel). The dashed line is a somewhat arbitrary threshold of “significant” LD, but according to this, we would say that LD is significant out to a distance of 1 million bases (1Mb). Therefore, if we have a significant marker, we should take a look at all the genes within a 1Mb radius of it. This sounds like a large distance, but keep in mind that it is 1/1000 the size of the 3B chromosome.

For some marker-trait associations, certain genes within our 1Mb window really stood out as likely candidates. For instance, one marker was highly associated with the traits grains per square meter, grains per head, and heads per square meter. This marker happened to be close to a gene that is similar to the rice gene abberant panicle organization 1 (APO1). As the name suggests, this gene is known to cause variation in rice panicle size and organization (the word “panicle” is used to refer to the rice equivalent of the “head” or “spike” in the wheat plant). Therefore, we had found a marker influencing traits related to the number of seeds per unit area, next to a gene similar to one known to affect the shape of rice plant heads. Among other candidate genes, we also found a heat-stress response gene close to a marker associated with the trait seeds per head. A previous study suggested that the expression of this gene is up-regulated during flowering and seed formation, so this will be an interesting finding to follow up on.

What’s next

While GWAS papers are great, we can’t draw many conclusions beyond the specific hypothesis that is being tested: does marker X have some association with trait Y, or no association? While locating nearby candidate genes is good practice, we still often don’t know the function of these genes. For that, we need follow-up studies. One brute-force method for deducing a gene’s function is to disable it completely (a “knock-out” mutation). You may have heard of CRISPR-Cas9 technology – this is one recently-developed way to introduce targeted mutations into a plant, including mutations that can effectively silence genes. Keep in mind that genes are variably expressed (i.e. translated into proteins) at different stages of development, so another brute force method of determining gene function is to put it under the control of a sequence that just turns up expression to 11 all the time. The hope is that some of these discoveries could in turn lead to future discoveries that could.. you know, save the planet and mankind. No pressure.

While taking this approach of identifying promising genes is all well and good, one problem is that many of these traits are highly polygenic (controlled by many genes). This means that any single gene likely only determines a small portion of the overall phenotypic variation that we observe. For highly polygenic traits, progress in breeding is mostly made by considering the genetic contributions of a vast set of genes as a whole. That is a topic to explore when we talk about the next paper.